Intervalley scattering by acoustic phonons in two-dimensional MoS2 revealed by double-resonance Raman spectroscopy.

ARTICLE Received 28 Oct 2016 | Accepted 23 Jan 2017 | Published 9 Mar 2017

DOI: 10.1038/ncomms14670

OPEN

Intervalley scattering by acoustic phonons in two-dimensional MoS2 revealed by double-resonance Raman spectroscopy Bruno R. Carvalho1,*, Yuanxi Wang2,*, Sandro Mignuzzi3,4, Debdulal Roy3,4, Mauricio Terrones2,5,6, Cristiano Fantini1, Vincent H. Crespi2, Leandro M. Malard1 & Marcos A. Pimenta1

Double-resonance Raman scattering is a sensitive probe to study the electron-phonon scattering pathways in crystals. For semiconducting two-dimensional transition-metal dichalcogenides, the double-resonance Raman process involves different valleys and phonons in the Brillouin zone, and it has not yet been fully understood. Here we present a multiple energy excitation Raman study in conjunction with density functional theory calculations that unveil the double-resonance Raman scattering process in monolayer and bulk MoS2. Results show that the frequency of some Raman features shifts when changing the excitation energy, and first-principle simulations confirm that such bands arise from distinct acoustic phonons, connecting different valley states. The double-resonance Raman process is affected by the indirect-to-direct bandgap transition, and a comparison of results in monolayer and bulk allows the assignment of each Raman feature near the M or K points of the Brillouin zone. Our work highlights the underlying physics of intervalley scattering of electrons by acoustic phonons, which is essential for valley depolarization in MoS2.

1 Departamento de Fı´sica, Universidade Federal de Minas Gerais, Belo Horizonte, Minas Gerais 30123-970, Brazil. 2 Department of Physics and Center for 2Dimensional and Layered Materials, The Pennsylvania State University, University Park, State College, Pennsylvania 16802, USA. 3 National Physical Laboratory, Hampton Road, Teddington TW11 0LW, UK. 4 Department of Physics, King’s College London, Strand, London WC2R 2LS, UK. 5 Department of Chemistry, The Pennsylvania State University, University Park, State College, Pennsylvania 16802, USA. 6 Department of Materials Science and Engineering, The Pennsylvania State University, University Park, State College, Pennsylvania 16802, USA. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed to B.R.C. (email: brunorc@fisica.ufmg.br) or to L.M.M. (email: lmalard@fisica.ufmg.br) or to M.A.P. (email: mpimenta@fisica.ufmg.br).

NATURE COMMUNICATIONS | 8:14670 | DOI: 10.1038/ncomms14670 | www.nature.com/naturecommunications

1

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14670

T

he second-order Raman spectrum of MoS2 and other semiconducting transition-metal dichalcogenides (TMDs) hosts a rich variety of features, which are strongly dependent on the number of layers and the excitation laser energy1–10. In the published works on two-dimensional (2D) MoS2, the second-order Raman bands were probed using only few laser excitation lines, and results were interpreted on the basis of phonon dispersion relations calculated from empirical force field models4. However, the use of multiple excitation energies is pivotal to unveiling the rich physical phenomena underlying the complex second-order Raman spectrum and the intervalley scattering processes in MoS2. Double-resonance Raman (DRR) is a special kind of secondorder process that involves the resonant scattering of excited electrons by phonons, and can be used to study electrons, phonons and their interplay11. By varying the incoming photon energy, the DRR condition selects different electronic states and different pairs of phonons with opposite finite momenta within the interior of the Brillouin zone (BZ)12–18. For graphene, the most important DRR features, the D and 2D bands, provide rich physical information about the sample12,19. In this work, we unravel the origin of the DRR processes occurring in MoS2 by investigating both experimentally and theoretically the Raman spectra as a function of the laser excitation. We measure the second-order bands using more than twenty different laser excitations ranging from 1.85 to 2.18 eV that densely cover the range of the A and B excitonic levels. Notably, we observe that the spectral positions of some specific second-order peaks depend on the laser excitation energy, which is characteristic of DRR processes. Results are explained by accurate first-principles calculations, which are shown to be crucial for identifying the different contributions to the DRR process. We show that some specific Raman processes are related to phonons near (but not at) the M and K points of the BZ. Moreover, the different contributions from the M phonons in monolayer and bulk reflect the crossover of the indirect-to-direct bandgap transition in the monolayer regime of MoS2. Our work provides a fundamental explanation of the resonant behaviour of the second-order Raman processes in MoS2, involving different electronic valleys, and may also be applicable to the higher-order Raman spectra and to other semiconducting TMDs. Results Experimental analysis. Figure 1a,b show the Raman spectra of monolayer (1L) and bulk MoS2 in the 350–500 cm 1 spectral range for three different laser energies (1.94, 2.04 and 2.11 eV). The two peaks around 388 and 407 cm 1 are associated, respectively, with the first-order in-plane and out-of-plane 1 for bulk) and A01 (A1g for bulk) Raman bands, with E0 (E2g 10,20 . All other features are contributions from symmetries different second-order processes. A first-order Raman band can be fitted with a Lorentzian curve, since it arises from a single phonon at the center of the BZ. On the other hand, a second-order Raman band is given by the convolution of multiple two-phonon processes across the whole BZ and, therefore, cannot be fitted by a sum of Lorentzian curves. The determination of the lineshape requires a complete theoretical description of Raman intensities, including electronphoton and electron-phonon coupling matrix elements. Nevertheless, the fitting of second-order bands in MoS2 and other TMD compounds by a sum of Lorentzian curves has been widely used in the literature3,4,7,10,21, because it provides a means to associate a feature in the spectrum with a specific phonon at a high-symmetry point within the BZ. Various works in the literature have fitted the second-order bands of MoS2 with different numbers of Lorentzians, and given the resulting peaks 2

different assignments (or when unassigned, different names)3,4,21. This fitting procedure can yield different peak numbers, positions, widths and intensities4,10,12,21. Here we also use this procedure, but we stress that it is only intended to provide a reliable estimate of the spectral position of the different contributions to the second-order Raman bands. The central goal of this work is to achieve a quantitative comparison between the measured and the calculated spectra within the entire spectral range considered here while making use of a dense sampling in terms of laser energies. We impose constant values for the full width at half maximum as a constraint in our Raman analysis, leaving intensity and position unconstrained. This procedure was adopted to decrease the number of fitting parameters, since the full width at half maximum is not expected to depend significantly on the laser energy within this narrow range of energy (1.85–2.18 eV). The number of Lorentzian peaks is increased until a pre-defined convergence threshold is reached, and the results of this procedure are confirmed by first-principles calculations, as described below. In Fig. 1a one observes a band at ca. 420 cm 1 that we call p1, and a broad and asymmetric band centered around 460 cm 1, which is commonly called the 2LA band in the literature. In this work, this broad band was fitted by four Lorentzian peaks: a first peak around 440 cm 1 and the other three denoted by p2, p3 and p4, as shown in Fig. 1a. The same number of peaks was used to fit the spectra of bulk MoS2, since more peaks were not found to be necessary to the fitting (in terms of convergence). This is possibly because the phonon branch splittings induced by interlayer interactions are not larger than B3 cm 1 in MoS2 (ref. 22). Figure 1c,d show the multiple excitation Raman map for 1L and bulk MoS2 obtained using more than twenty laser lines with energies from 1.85 to 2.18 eV. The horizontal scale represents the Raman shift, while the vertical scale is the laser excitation energy. We can observe the resonances of all Raman bands across the A (B1.89 eV) and B (B2.06 eV) excitonic transitions, which are marked by horizontal dashed lines. An important result is the dispersion of some Raman features as the laser energy changes, as clearly revealed by the dashed lines tracking Raman peak positions (Fig. 1a–d). Each second-order band in the range 400–480 cm 1, measured with different laser excitation energies, was fitted according to the same procedure used in Fig. 1a,b. The positions of p1, p2, p3 and p4 as a function of the laser excitation energy are plotted in Fig. 1e,f for 1L and bulk MoS2, respectively. Notably, the position of the p2 peak is unchanged for all laser energies, whereas p1, p3 and p4 red-shift as the laser excitation energy increases. Theoretical model. To explain the experimental results, we calculated the second-order Raman spectra for 1L and bulk MoS2 within the single-particle picture, using the electronic structure and phonon dispersion obtained from density functional theory (DFT) (see Supplementary Note 1 and Supplementary Fig. 1) for different laser excitation energies, as described in Theoretical methods. Figure 2a shows a schematic representation of a DRR process in MoS2, where the low-energy electronic structure is represented by parabolic bands at the BZ edges. The DRR process begins with an incoming photon creating an electron-hole pair of wave vector k near the K valley. The electron is then inelastically scattered by the emission of a phonon with branch index m, wave vector q, and energy ‘ om q to the K0 valley. After that, the electron is inelastically scattered back to the K valley by the emission of a second phonon with branch index n, wave vector q and energy ‘ onq , where the electron-hole pair recombines emitting a photon with energy EL ‘ om q ‘ onq (Stokes scattering). At most


ARTICLE


b

1L-MoS2 2.11 eV

A′1(Γ)

Intensity (a.u.)

E ′(Γ)

p2

p4 E 12g(Γ)

A1g(Γ)+E 22g(Γ)

2.04 eV

1.94 eV

f

1L-MoS2

A1g(Γ)

p3

p1

e

Bulk

Raman shift (cm–1)

a

Bulk

472 468 464 460 456 452 p1 peak p2 peak p3 peak p4 peak Theory

424 420 416

360

400

440

480

360


400

440

412


1.89 1.96 2.03 2.10 2.17 1.89 1.96 2.03 2.10 2.17 Laser energy (eV)

d 2.17

p1

p2

p1

p4

p2

p3

p4

g p3

p2 2

B

Intensity (a.u.)

2.10 B

2.03

1.96

1.89

A

360

A

400

440


480

360

440

240

p4 2 1.92 eV

200 400

h

p3 2

220 240 260 280 Raman shift (cm–1)


c

Laser energy (eV)

480

237

p2/2 peak p3/2 peak p4/2 peak Theory

234 231 228 225

1.89 1.96 2.03 2.10 2.17 Laser energy (eV)

480


Figure 1 | Resonance Raman results on 1L and bulk MoS2. (a,b) Raman spectra of 1L and bulk MoS2 measured with three laser excitation energies: 1.94, 2.04 and 2.11 eV. The spectra are fit to a sum of Lorentzians (grain and green curves). The second-order bands studied here are denoted p1, p2, p3 and p4. (c,d) Resonant Raman maps of 1L and bulk MoS2 acquired using more than twenty different laser lines, showing the enhancement of the Raman bands across the A and B excitons (horizontal dashed lines). (e,f) Laser energy dependence of the experimental values of the positions of the p1, p2, p3 and p4 peaks (symbols) of 1L and bulk MoS2, and the calculated dispersion of these peaks using equation (1) (solid curves). (g) Raman spectrum of a defective 1L MoS2 sample, showing the disorder-induced bands associated with acoustic phonons near the edges of the Brillouin zone. Defects were created through bombardment with Mn þ (see Experimental Methods section for details). The band was fit to a sum of Lorentzians with frequencies half the frequencies of p2, p3 and p4. (h) Laser energy dependence of the experimental values of p2/2, p3/2 and p4/2 (symbols) showed in g, and the calculated dispersion of these disorder-induced peaks (solid curves) assuming one-phonon-defect DRR scattering. The absence of the theoretical curve of p4 peak in e,h is due to its weak intensity not observable in the calculated Raman spectra (see Fig. 3 and, main text for details). The error bars in e,f,h represent the standard error from the fitting process.

two of these steps can be simultaneously resonant for a DRR process9,11–13. The intensity of a second-order Raman process is given by following expression: 2 X Mf Mcb Mba M0 pp Iee ðEL Þ¼ g g g ; c v m n EL Ekc þ q þ Ekv ‘ om q i EL Ekc þ Ekv i k;q;m;n EL Ek þ Ek ‘ o q ‘ oq i 2 2 2

ð1Þ where M0, Mf are the matrix elements of the exciton–photon interactions for the incoming and outgoing photons, Mcb , Mba represent the exciton–phonon interactions23, q and k are the wave vectors of the phonon and the electron, respectively, Ekc þ q and Ekc are the energies of the intermediate states, and m and n denote the phonon branches involved in the process. The damping constant g is related to the finite lifetime of the intermediate states, and ‘ om;n q is the corresponding phonon energy. In the present study, we only focus on two-electron processes (ee) involving transverse acoustic (TA) and longitudinal acoustic (LA) phonons; processes involving hole scattering (hh, eh and he)24 were not taken into account for clarity and will be elaborated in future studies. The matrix elements are taken to be constants in the calculations.

Following the usual analysis for DRR in graphite25, the calculated DRR intensities are dominated by contributions from phonon wave vectors near qB0 and qB2k (distance from the K point), as shown in Supplementary Fig. 1. As the laser energy increases, the peak corresponding to qB0 is almost dispersionless and the peak corresponding to qB2k is dispersive25. It has been reported for graphene26–28 that the qB0 contribution vanishes due to the symmetry selections rules and quantum interference. We show later that, in the case of MoS2, this contribution, which is present in the DFT results, should vanish if excitonic effects are considered. For now, we only focus on the qB2k contribution by tracking the most dispersive Raman peaks in the calculated DRR intensities, since the qB0 contributions are almost dispersionless (other dispersionless features such as van Hove singularities can be identified separately from the phonon density of states (pDOS), as shown later). Figure 2b,c show a contour map of the conduction electron energies for 1L and bulk MoS2, where the conduction band minimum occurs at K for 1L MoS2. Based on the calculated electronic structure, Fig. 2d shows three sets of electronic states that participate in the DRR process in 1L MoS2, for laser excitation energies of 1.9, 2.0 and 2.1 eV, represented,


3

ARTICLE


a

b

c

1L-MoS2

Bulk

– 2h q

q

1.0

E ck+q

–q EL

0.8

K

K′ 0.6

Q K′

E vk

Γ

0.4 0.2

M

K

d

K

K′

Energy above CBM (eV)

K

k E ck

0

e

K

f

K

Q

Γ

Q

K

Γ

K

Q

M M

M

M

Figure 2 | Double-resonance Raman model. (a) Representation of a DRR process where excited electronic states are connected by two phonons with opposite momenta q and q. (b,c) Plot of the conduction bands for 1L and bulk MoS2, respectively. (d) Locus of electronic states that participate in the DRR Raman process around K for laser energies of 1.9, 2.0 and 2.1 eV in red, green and blue. Possible phonon wave vectors that connect these states with each other and with other high-symmetry points (M and Q) are shown as red dashed arrows. (e,f) Density of states of phonons that satisfy the DRR conditions for 1L and bulk MoS2, respectively. The DRR intensity in e,f is in arbitrary units justifying the use of no colour bars.

respectively, in red, green and blue. Similar to the case of graphene19, when the laser excitation energy increases the locus of the on-resonance electronic states expands outward from K. Figure 2d also shows that phonons that connect the pockets around K and K0 valleys have wave vectors near K, denoted qBk here (the symbol ‘B’ means in the vicinity of the given highsymmetry point). On the other hand, pockets around the K and Q points are connected by phonons near both the M and Q points, with wave vectors qBM and qBQ, where Q is another conduction band local minimum half-way between C and K, as shown in Fig. 2b–d. Figure 2e shows the on-resonance density of states for acoustic phonons that connect two pockets in the electronic structure at 2.0 eV excitation for 1L MoS2. A high density of qBk phonons appears around K and contributes strongly to the DRR scattering. A weaker density of phonons around M also appears (green arrow in Fig. 2e). They originate from scattering between the K and Q valleys (see Fig. 2d). For 1L MoS2, the intervalley scattering between K and Q only contributes weakly to the DRR process since the two minima are misaligned by 0.2 eV for 1L MoS2. The above procedure is also applied to bulk MoS2, where the conduction band local minimum at Q drops in energy, rendering the band gap indirect29,30, as shown in Fig. 2c. Now, both qBk and qBM phonons in bulk MoS2 contribute strongly to the on-resonance pDOS (red and green arrows in Fig. 2f), as well as the additional scattering channel between K and Q with phonon wave vector qBQ (see Fig. 2d). Figure 3a–d show the calculated second-order Raman spectra for 1L and bulk MoS2 for different laser energies, as given by equation (1), and using the phonon dispersion relations of the LA and TA phonons (Supplementary Fig. 1). In the range 405–420 cm 1 (Fig. 3a,c), we observe one dispersive and one 4

non-dispersive feature, and in the range 450–470 cm 1 (Fig. 3b,d), one non-dispersive and two dispersive features at higher frequency. A normal second-order Raman band is non-dispersive, whereas a dispersive behaviour indicates a DRR process. The positions of each feature in Fig. 3a–d as a function of the laser energy are also plotted in Fig. 1e,f, for 1L MoS2 and bulk, respectively, and represented by the black lines. The comparison between the experimental and calculated results in Fig. 1e,f shows an excellent agreement, and will allow for the assignment of each second-order feature. As qBk expands away from K, the frequency of on-resonance LA phonons would decrease with increasing laser energy due to the negative curvature of the LA branch dispersion near K. Although p4 is not clearly observable in the calculated DRR spectra for 1L MoS2, a small contribution near BM can be seen in the BZ mapping of DRR intensity in Fig. 2e (green arrow). The underestimated intensity of this contribution from M, compared with the experimental results (where the contribution from M appears larger), might be due to strong electron-phonon coupling matrix elements between BK and BQ conduction valleys by an qBM phonon27 (matrix elements are treated as constants in this study). Assignment of the second-order Raman features of MoS2. The 2LA band. The strongest feature in the second-order spectrum of MoS2 is the broad and asymmetric band centered around 460 cm 1. In previous Raman studies of bulk MoS2 (ref. 1), it was initially ascribed to the overtone of the LA phonon at the M point in the BZ and assigned to the 2LA(M) band1,2. However, its asymmetric shape led some authors to suggest that it could have contributions from other second-order processes. Verble and Wieting6 assigned the shoulder of the 2LA band at


ARTICLE


a

c

b

1L-MoS2 p2

p1

p1

p3

d

Bulk

p2

p4

p3

2.16 2.10 2.07 2.04 2.01 1.98

Laser energy (eV)

DR intensity (a.u.)

DR intensity (a.u.)

2.13

1.95 1.92

415

420

425 455 460 465 470

415


420

425

455

465


e

f K

TA

K

LA

0 –5

–10

M

M

Frequency –1 change (cm )

410

–15

Figure 3 | Calculated resonance Raman spectra. Calculated second-order Raman for (a,b) 1L and (c,d) bulk MoS2 for different laser energies, using equation (1). All spectra are vertically shifted for clarity. (e,f) Calculated phonon frequencies of the TA and LA branches with respect to the frequency maxima of each branch, showing that the LA branch disperses more rapidly away from K.

466 cm 1 to the Raman-inactive mode with A002u symmetry (or A2u mode for bulk or even layers) at the C point. Gołasa et al.4 suggested this band was composed by a combination of the optical E1g phonon with the TA phonon at the M point. Livneh and Spanier21 proposed different contributions for the 2LA band in bulk MoS2, such as the pDOS, overtones of the M edge phonons from the acoustic (LA) and quasi-acoustic (LA0 ) branches, which are very close in energy, and a weak contribution of the LA(K) phonons. In this work, we assign the different contributions to the 2LA band by comparing the experimental and calculated multiple excitation Raman spectra of both 1L and bulk MoS2. In our analysis, the 2LA band was fitted by four peaks, a peak around 440 cm 1 and the peaks p2, p3 and p4, as shown in Fig. 1a,b. The non-dispersive behaviour of p2 (see Fig. 1e,f) shows that it corresponds to a normal second-order Raman process. On the other hand, p3 and p4 exhibit a dispersive behaviour, a signature of a DRR process (or higher-order resonant process) due to photons with different energies selecting electrons and phonons with different wave vectors in the BZ. Our calculations show that p2 comes from the van Hove singularity (vHs) in the pDOS from a saddle point between K and M. The frequency of p2 matches with twice the calculated frequency of the vHs. Our assignment for p2 agrees with the conclusion reached by Livneh and Spainer21 (the L2 peak in their work). The 440 cm 1 peak was 2 (ref. 21). recently ascribed to the combination mode A1g þ E2g However, it can also be due to the asymmetric shape of p2, since the vHs is not necessarily symmetric. We next discuss the origin of the p3 and p4 peaks. Previous works assigned them to 2LA(M) and/or 2LA(K) processes, but the near-degeneracy of the LA phonon at M and K prevented a clear distinction between these two contributions1–5,31. A comparison of our results for 1L and bulk MoS2 allows for an unambiguous assignment. As shown in Fig. 2f, the DRR process in bulk MoS2 can support scattering by both qBM and qBK phonons, whereas DRR in 1L MoS2 (Fig. 2e) is very weak

for qBM phonons. The different DRR contributions from BM phonons in 1L and bulk MoS2 allow us to distinguish the contributions of DRR BK and BM phonons to the second-order Raman band, as will be shown below. The calculated results in Fig. 3b,d show that p4 clearly appears in the spectra of bulk MoS2, but is very weak for 1L MoS2. Notice that different from the calculated results, the experimental intensity of p4 for 1L MoS2 is weak but not negligible, this is possibly ascribed to the effect of a strong electron-phonon coupling of p4. However, the most relevant result is that p4 is more intense than p2 and p3 for 2L, 3L and bulk MoS2 (see Supplementary Note 2 and Supplementary Fig. 2), and weaker for 1L MoS2. We, therefore, conclude that the p4 peak is related to the scattering of the excited electron between the K and the Q valleys by two LA phonons near BM (see Fig. 2d). This leaves p3 to be assigned to the scattering process between K and K0 by two LA phonons in the vicinity of the K point. Thus, a more precise assignment for p3 and p4 would be 2LA(BK) and 2LA(BM), respectively. Figure 1e,f compare the experimental positions of p3 and p4 with their calculated dispersion. The excellent agreement between experiment and theory further confirms the assignment of p3 and p4 as 2LA(BK) and 2LA(BM). As the laser energy increases, p3 and p4 disperse at rates of 49 and 21 cm 1 eV 1, respectively, and this result reflects the different slopes of the LA phonon dispersion near K and M of MoS2. We thus conclude that the broad and asymmetric band centered around 460 cm 1 can be explained by contributions from LA phonons, instead of optical phonons4,6. It has contributions from LA phonons near the saddle point between K and M (pDOS singularities) and from LA phonons near BK and BM, which are enhanced in the spectra by the intervalley DRR process. The p1 (or b) peak. In a resonance Raman study of bulk MoS2 at 7 K, Sekine et al.32 observed a band around 430 cm 1 with a dispersive behaviour near the A and B excitonic resonances, and called it the b band (see Supplementary Note 3 and


5

ARTICLE


Supplementary Fig. 3). This band was ascribed to a two-phonon process involving a quasi-acoustic breathing mode and 2interlayer a transverse optical phonon E1u both along the c-axis32. However, this band was also observed in more recent Raman studies of monolayer MoS2 (ref. 21), thus ruling out the assignment suggested by Sekine et al.32. In a detailed multiphonon study of bulk MoS2, Livneh and Spanier suggested that the b band could involve combinations of the LA (or LA0 ) and TA (or TA0 ) phonons at the K point21. In Fig. 1e,f we compare the experimental values of the p1 positions (red circles) with the dispersion of the calculated peak (black curve) considering an intervalley DRR process involving one LA phonon and one TA phonon, in both time orders, in the vicinity of K (see Fig. 3a,c). The combined energy of the relevant LA þ TA phonons near BK red-shifts as the incident laser energy increases, dispersing at a rate of 43 cm 1 eV 1. The agreement between the experimental and calculated results in Fig. 1e,f allows for the assignment of p1 to LA(BK) þ TA(BK). The larger dispersion of p3 ¼ 2LA(BK) than p1 ¼ LA(BK) þ TA(BK) is consistent with TA phonons dispersing more slowly near K than do LA phonons. This can be seen in Fig. 3e,f, which show the phonon dispersion for the TA and LA branches in the irreducible BZ, where the frequencies are represented by a colour map. The LA phonon frequency rapidly decreases away from K, whereas the TA phonon frequency varies slowly near K. The 2LA(BQ) band. As represented in Fig. 2d, there are two different scattering processes between the K and Q electronic valleys, involving phonons with both wave vectors qBM and qBQ. Indeed, the theoretical DRR maps in Fig. 2f predict the existence of a 2LA(BQ) band for bulk MoS2 (see the red circle labelled BQ in Fig. 2f) at a frequency of 370 cm 1. This band was not detected in our room-temperature spectra, presumably because the electron-phonon coupling matrix elements for this process are almost vanishing27. However, it was possibly observed in previous low-temperature studies in bulk MoS2 (refs 21,32). Sekine et al.32 observed a weak feature around 380 cm 1 in the spectra at 7 K, and ascribed it to the zone-center Raman-inactive 1 phonon. Livneh and Spanier reported the existence of a band E2u at 386 cm 1, observed at 95 K, which could not be explained by their multi-phonon analysis21. A multiple excitation Raman study at low temperatures will be necessary to confirm the existence of the double-resonance 2LA(BQ) band, and its possible dispersive behaviour. Disorder-induced DRR bands. As a further verification of our proposed DRR conclusions involving the acoustic phonons, the defect-induced resonant Raman spectra in 1L MoS2 were measured. Figure 1g shows the spectrum in the range 200–285 cm 1, where we can observe defect-induced Raman features33 for a excitation energy of 1.92 eV (see also Supplementary Note 4 and Supplementary Fig. 4). This process is similar to the disorderinduced D band in graphene19, which also comes from a DRR process involving just one phonon, where momentum conservation is provided by elastic scattering of the excited electron by a defect9,12,19. We have fitted the defect-induced Raman band with four Lorentzian peaks, similar to the procedure used for the 2LA second-order band in 1L pristine MoS2. The frequencies of the three peaks at 227, 234 and 235 cm 1 correspond, respectively, to one-half the frequencies of the p2, p3 and p4 peaks as marked by the vertical dashed lines in Fig. 1g. Figure 1h shows the frequencies of these three Raman peaks as a function of the laser excitation energy, and the calculated dispersion of the DRR features considering now scattering by a phonon and a defect. The excellent agreement with the theory and experiment further supports the interpretation of 6

the second-order DRR features, and highlights the intervalley elastic scattering by defects. Excitonic effects. We now discuss the validity of the singleparticle picture in the presence of excitonic effects. Single-particle DFT calculations have been extensively used in studying DRR processes in graphene, where excitonic effects are negligible at low excitation energies34. The strong excitonic effects in 1L MoS2 call for further interrogation of assignments based on singleparticle DFT results. Here based on a simple two-band model, we show that the use of single-particle band structures (see Supplementary Note 5 and Supplementary Fig. 5a,c,e) and exciton band structures (see Supplementary Note 5 and Supplementary Fig. 5b,d,f) would yield similar laser-energy dependence of the Raman frequencies, thereby justifying the use of DFT eigenvalues in equation (1). In the single-particle case, we model the low-energy band structure as parabolic bands E± ¼ ±(Eg/2 þ k2/2m*), where for convenience we assume that me ¼ mhm* is the mass of the electron and hole and that the conduction band has been rigidly shifted to bring the single-particle band-gap to match with the optical gap Eg. The actual electron-hole mass ratio calculated from different first-principles methods ranges from 0.8 to 1.8 (refs 35–38). From the single-particle bands, excitons can be approximately represented by pairing electrons and holes with matching group velocities at ke ¼ kh k (where k is the wave vector measured from K); such an exciton has a center-of-mass momentum of Q¼2k and a total energy of Eg þ 2 k2 =2m ¼Eg þ Q2 =2M. Thus, we find M ¼ 2m*, that is, the excitonic dispersion has half the band curvature of the single-particle bands. This is consistent with the semi-classical interpretation that the mass of an exciton is the sum of its constituents’ masses. Momentum-resolved electron energy-loss spectroscopy measurements39 are consistent with the exciton mass being larger than the constituent hole and electron masses obtained from first-principles calculations: 0.9–1.4 times their sum, depending on the exchange-correlation functional and pseudopotential used (note that the discussion below still applies, semi-quantitatively, for a reasonable range of exciton masses). Finally, we assume that the excitonic dispersion near Q¼0 and Q¼ K can be described by parabolic bands with the same band minimum and same curvature, since they both describe electron-hole pairs near the two degenerate valleys. That is, the energy difference of the two minima calculated by solving the tight-binding-based Bethe-Salpeter equation, 0.015 eV, is neglected here40. We now discuss intervalley scattering between single-particle states and between excitonic states within the above assumptions, as depicted in Supplementary Fig. 5c,d. When the laser energy is EL ¼ Eg þ 2k2/2m*, the on-resonance phonons in the singleparticle bands are 2k away from K, as previously described. When the excitonic dispersion replaces the single-particle energies in equation (1), the resonance condition will be satisfied only if the incoming photon matches the optical band gap, that is, EL ¼ Eg, where excitons are scattered to Q¼K and back. However, as the laser energy increases beyond the optical gap EL4Eg (but still within the exciton linewidth), a single and weaker resonance can still be achieved when the exciton is scattered to the other valley near Q¼K, and then back. The same laser energy as the one given in the single-particle system can be recast in terms of the exciton mass EL ¼ Eg þ 2k2/2m* ¼ Eg þ (2k)2/2M. Hence the onresonance phonons scattering excitons are also away from K by 2k, the same as the single-particle case. Therefore, the excitationenergy dependence of the Raman frequencies (but not the intensities) can be reasonably well described by a single-particle


ARTICLE


electronic band structure, as both pictures give the so-called qB2k contribution of on-resonance phonons. We note that the qB0 contribution, which appears in the single-particle picture, is lost in the excitonic picture. The single-particle results may also carry over to the manybody picture due to the presence of trions. Unlike neutral excitons with vanishing momenta, trions of finite momenta can be created where the momentum comes from the extra charge carrier. Thus, the momentum of this many-body state again depends on the laser excitation energy, and a dispersive behaviour is expected. Our conclusion that the LA(K) phonon dominates the intervalley scattering is consistent with previous reports of valley depolarization in TMDs41–43. From the decay of the valley polarization with increasing temperature, Zeng et al.41 extracted the energy of the phonon mode dominating intervalley scattering to be 240 cm 1, consistent with the LA(K) phonon energy. However, this thermal destruction of valley polarization could also originate from the excitonic levels shifting as a function of temperature, thus detuning the laser excitation energy from the A exciton energy42. To rule out this possibility, Kioseoglou et al.43 performed depolarization measurements as a function of increasing pump laser energy at T ¼ 5 K, whereby the excess energy enables phonon-assisted intervalley scattering. These authors found that the destruction of the valley polarization occurs beyond an excess energy of 60 meV (E480 cm 1), which corresponds to two-LA(K) phonons, thus showing that intervalley scattering by LA phonons is responsible for depolarization43. Discussion To summarize, this work explains the origin of the DRR process in monolayer and bulk MoS2, as involving different intervalley scattering processes. The resonant Raman spectra of the most intense second-order features in MoS2 and the associated firstorder disorder-induced bands, were measured using many laser excitation energies in the range of 1.85–2.18 eV, which covers the A and B excitonic levels. Experimental results are explained by DFT calculations of the DRR scattering in monolayer and bulk MoS2. We demonstrate that the use of multiple excitation energies is crucial for understanding the physical phenomena underlying the second-order Raman spectrum of MoS2. We observe that the spectral position of some specific secondorder peaks depends on the laser excitation energy, which is characteristic of DRR processes. By varying the incoming photon energy, the DRR condition selects different electronic states in the K and Q valleys, and different pairs of phonons with opposite finite momenta near (but not at) the M and K points of the BZ. Our results show that the DRR process reflects the indirect-todirect bandgap transition from bulk to monolayer, and this effect allows the assignment of the Raman features to specific phonons near M or K (see also Supplementary Note 6 and Supplementary Tables 1–3). Our study can also be extended to explain the second-order Raman spectra and the double-resonance process in other semiconducting TMDs, such as MoSe2, WS2, WSe2. Moreover, the methodology in this work, based on multiple excitation Raman results and first-principle calculations, can also be used to explain the multiphonon spectra of semiconducting TMDs that exhibit a rich variety of high-frequency features, up to the fifth-order4,21,44. Finally, we show the second-order DRR spectra of MoS2 originates in intervalley scattering by acoustic phonons, a mechanism which is also responsible for the destruction of valley polarization (that is, depolarization)41,43. Our work is thus relevant for the field of valleytronics of MoS2, since the robustness of valley polarization depends sensitively on the absence of intervalley scattering41,42,45.

Methods Experimental methods. The MoS2 samples were obtained by mechanical exfoliation of natural 2H-MoS2 crystals transferred onto Si substrates with a 298 nm thick SiO2 coating. Defective monolayer MoS2 was created through bombardment with Mn þ in an ultrahigh vacuum time-of-flight secondary ion mass spectrometry (TOF-SIMS IV) instrument (ION-TOF GmbH, Muenster, Germany), equipped with a liquid metal ion gun at an angle of 45° to the surface normal and using an ion-beam kinetic energy of 25 keV, as described in ref. 33. The ion current and the exposure time were tuned to obtain an average inter-defect distance of 2.2 nm. Micro-Raman measurements were obtained on a DILOR XY triple-monochromator spectrometer equipped with a N2-cooled charge-couple device detectors and with 1,800 g mm 1 diffraction gratings, giving spectral resolution better than 1 cm 1. The MoS2 samples were excited using different laser sources (Ar/Kr, and dye laser with DCM special and rhodamine 6G) covering excitation energies from 1.85 to 2.18 eV. All measurements were conducted at room temperature in backscattering geometry where the laser propagation is perpendicular to the MoS2 layer plane. A 100 objective provided a spot of 1 mm diameter. The laser power at the sample surface was kept below 1.0 mW to avoid sample heating. The Raman spectra were normalized by the intensity of the Si peak at each excitation energy and, the Si Raman peak at 521.6 cm 1 is used to calibrate the Raman shift. Theoretical methods. DFT calculations were performed using the generalized gradient approximation parametrized by Perdew-Burke-Ernzerhof for the exchange correlation functional46. Electron–nucleus interactions are described within the projector augmented wave formalism47,48. All structural relaxation and band structure calculations were carried out by the Vienna Ab-initio Simulation Package49,50, with energies converged at a plane wave expansion energy cutoff of 400 eV and forces converged at 0.004 eV Å 1. Interatomic force constants were calculated using the supercell method with the size of the supercell converged at 6 6 1. Interlayer van der Waals interactions were included using the semiempirical DFT-D2 method51. All conduction bands calculated from DFT have been rigidly shifted to match with the experimentally measured optical gaps for 1L and bulk MoS2. Resonant Raman calculations were performed using phonon dispersions sampled on a 400 400 1 C-centered grid for 1L MoS2 and a 200 200 50 grid for bulk. A phenomenological damping constant (smearing) of g ¼ 0.02 eV was used. Since the absolute Raman shifts are prone to slight (1%) over/under-estimates due to limitations of DFT, the calculated phonon frequencies were shifted by within 1% to allow better comparison with the respective experimental dispersions. This shift leaves the calculated dispersion in the Raman shift with laser energy (which reflects the influence of the phonon and electron band structures on the DRR process) unchanged. Data availability. All relevant data are available from the authors upon request.

References 1. Chen, J. & Wang, C. Second order Raman spectrum of MoS2. Solid State Commun. 14, 857–860 (1974). 2. Stacy, A. & Hodul, D. Raman spectra of IVB and VIB transition metal disulfides using laser energies near the absorption edges. J. Phys. Chem. Solids 46, 405–409 (1985). 3. Frey, G. L., Tenne, R., Matthews, M. J., Dresselhaus, M. S. & Dresselhaus, G. Raman and resonance Raman investigation of MoS2 nanoparticles. Phys. Rev. B 60, 2883–2892 (1999). 4. Gołasa, K. et al. Multiphonon resonant Raman scattering in MoS2. Appl. Phys. Lett. 104, 092106 (2014). 5. Liu, H.-L. et al. Anomalous lattice vibrations of monolayer MoS2 probed by ultraviolet Raman scattering. Phys. Chem. Chem. Phys. 17, 14561–14568 (2015). 6. Verble, J. L. & Wieting, T. J. Lattice mode degeneracy in MoS2 and other layer compounds. Phys. Rev. Lett. 25, 362–365 (1970). 7. Lee, C. et al. Anomalous lattice vibrations of single- and few-layer MoS2. ACS Nano 4, 2695–2700 (2010). 8. Chakraborty, B., Matte, H. S. S. R., Sood, A. K. & Rao, C. N. R. Layer-dependent resonant Raman scattering of a few layer MoS2. J. Raman Spec. 44, 92–96 (2013). 9. Pimenta, M. A., del Corro, E., Carvalho, B. R., Fantini, C. & Malard, L. M. Comparative study of Raman spectroscopy in graphene and MoS2-type transition metal dichalcogenides. Acc. Chem. Res. 48, 41–47 (2015). 10. Carvalho, B. R., Malard, L. M., Alves, J. M., Fantini, C. & Pimenta, M. A. Symmetry-dependent exciton-phonon coupling in 2D and bulk MoS2 observed by resonance Raman scattering. Phys. Rev. Lett. 114, 136403 (2015). 11. Martin, R. M. & Falicov, L. M. Resonant Raman Scattering. in Light Scattering in Solids I. (ed. Cardona, M.) Vol. 8, 79–145 (Springer, Berlin, Germany, 1983). 12. Malard, L. M., Pimenta, M. A., Dresselhaus, G. & Dresselhaus, M. Raman spectroscopy in graphene. Phys. Rep. 473, 51–87 (2009). 13. Maultzsch, J., Reich, S. & Thomsen, C. Double-resonant Raman scattering in graphite: interference effects, selection rules, and phonon dispersion. Phys. Rev. B 70, 155403 (2004).


7

ARTICLE


14. Carvalho, B. R. et al. Probing carbon isotope effects on the Raman spectra of graphene with different 13C concentrations. Phys. Rev. B 92, 125406 (2015). 15. Mohr, M., Machoń, M., Maultzsch, J. & Thomsen, C. Double-resonant Raman processes in germanium: Group theory and ab initio calculations. Phys. Rev. B 73, 035217 (2006). 16. Bansal, M. L., Sood, A. K. & Cardona, M. Strongly dispersive low frequency Raman modes in germanium. Solid State Commun. 78, 579–582 (1991). 17. Renucci, M. A., Renucci, J. B., Zeyher, R. & Cardona, M. Second-order Raman scattering in germanium in the vicinity of the E1, E1 þ D1 edges. Phys. Rev. B 10, 4309–4323 (1974). 18. Cerdeira, F., Dreybrodt, W. & Cardona, M. Resonant Raman scattering in germanium. Solid State Commun. 10, 591–595 (1972). 19. Pimenta, M. A. et al. Studying disorder in graphite-based systems by Raman spectroscopy. Phys. Chem. Chem. Phys. 9, 1276–1290 (2007). 20. Ribeiro-Soares, J. et al. Group theory analysis of phonons in two-dimensional transition metal dichalcogenides. Phys. Rev. B 90, 115438 (2014). 21. Livneh, T. & Spanier, J. E. A comprehensive multiphonon spectral analysis in MoS2. 2D Mater. 2, 035003 (2015). 22. Molina-Sańchez, A. & Wirtz, L. Phonons in single-layer and few-layer MoS2 and WS2. Phys. Rev. B 84, 155413 (2011). 23. Garcıá-Cristo´bal, A., Cantarero, A., Trallero-Giner, C. & Cardona, M. Excitonic model for second-order resonant Raman scattering. Phys. Rev. B 49, 13430–13445 (1994). 24. Venezuela, P., Lazzeri, M. & Mauri, F. Phonons in single-layer and few-layer MoS2 and WS2. Phys. Rev. B 84, 035433 (2011). 25. Cançado, L. G. et al. Stokes and anti-stokes double resonance Raman scattering in two-dimensional graphite. Phys. Rev. B 66, 035415 (2002). 26. Jin, Z., Li, X., Mullen, J. T. & Kim, K. W. Intrinsic transport properties of electrons and holes in monolayer transition-metal dichalcogenides. Phys. Rev. B 90, 045422 (2014). 27. Li, X. et al. Intrinsic electrical transport properties of monolayer silicene and MoS2 from first principles. Phys. Rev. B 87, 115418 (2013). 28. Kaasbjerg, K., Thygesen, K. S. & Jacobsen, K. W. Phonon-limited mobility in n-type single-layer MoS2 from first principles. Phys. Rev. B 85, 115317 (2012). 29. Mak, K., Lee, C., Hone, J., Shan, J. & Heinz, T. Atomically thin MoS2: A new direct-gap semiconductor. Phys. Rev. Lett. 105, 136805 (2010). 30. Kormańyos, A. et al. Trigonal warping, the C valley, and spin-orbit coupling effects. Phys. Rev. B 88, 045416 (2013). 31. Lee, J.-U., Park, J., Son, Y.-W. & Cheong, H. Anomalous excitonic resonance Raman effects in few-layer MoS2. Nanoscale 7, 3229–3236 (2015). 32. Sekine, T., Uchinokura, K., Nakashizu, T., Matsuura, E. & Yoshizaki, R. Dispersive Raman mode of layered compound 2H-MoS2 under the resonant condition. J. Phys. Soc. Jpn 53, 811–818 (1984). 33. Mignuzzi, S. et al. Effect of disorder on Raman scattering of single-layer MoS2. Phys. Rev. B 91, 195411 (2015). 34. Yang, L., Deslippe, J., Park, C.-H., Cohen, M. L. & Louie, S. G. Excitonic effects on the optical response of graphene and bilayer graphene. Phys. Rev. Lett. 103, 186802 (2009). 35. Qiu, D. Y., da Jornada, F. H. & Louie, S. G. Optical spectrum of MoS2: manybody effects and diversity of exciton states. Phys. Rev. Lett. 111, 216805 (2013). 36. Ramasubramaniam, A. Large excitonic effects in monolayers of molybdenum and tungsten dichalcogenides. Phys. Rev. B 86, 115409 (2012). 37. Cheiwchanchamnangij, T. & Lambrecht, W. R. L. Quasiparticle band structure calculation of monolayer, bilayer, and bulk MoS2. Phys. Rev. B 85, 205302 (2012). 38. Shi, H., Pan, H., Zhang, Y.-W. & Yakobson, B. I. Quasiparticle band structures and optical properties of strained monolayer MoS2 and WS2. Phys. Rev. B 87, 155304 (2013). 39. Habenicht, C., Knupfer, M. & Bu¨chner, B. Investigation of the dispersion and the effective masses of excitons in bulk 2H-Mos2 using transition electron energy-loss spectroscopy. Phys. Rev. B 91, 245203 (2015). 40. Wu, F., Qu, F. & MacDonald, A. H. Exciton band structure of monolayer MoS2. Phys. Rev. B 91, 075310 (2015). 41. Zeng, H., Dai, J., Yao, W., Xiao, D. & Cui, X. Valley polarization in MoS2 monolayers by optical pumping. Nat. Nanotechnol. 7, 490–493 (2012). 42. Mak, K. F., He, K., Shan, J. & Heinz, T. F. Control of valley polarization in monolayer MoS2 by optical helicity. Nat. Nanotechnol. 7, 494–498 (2012). 43. Kioseoglou, G., Hanbicki, A. T., Currie, M., Friedman, A. L. & Jonker, B. T. Optical polarization and intervalley scattering in single layers of MoS2 and MoSe2. Sci. Rep. 6, 25041 (2016).

8

44. Berkdemir, A. et al. Identification of individual and few layers of WS2 using Raman spectroscopy. Sci. Rep. 3, 1755 (2013). 45. Xiao, D., Liu, G.-B., Feng, W., Xu, X. & Yao, W. Coupled spin and valley physics in monolayers of MoS2 and other group-VI dichalcogenides. Phys. Rev. Lett. 108, 196802 (2012). 46. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865–3868 (1996). 47. Blo¨chl, P. E. Projector augmented-wave method. Phys. Rev. B 50, 17953–17979 (1994). 48. Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 59, 1758–1775 (1999). 49. Kresse, G. & Furthmu¨ller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169–11186 (1996). 50. Kresse, G. & Furthmu¨ller, J. Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set. Comput. Mater. Sci. 6, 15–50 (1996). 51. Grimme, S. Semiempirical GGA-type density functional constructed with a long-range dispersion correction. J. Comput. Chem. 27, 1787–1799 (2006).

Acknowledgements We thank Profs Tony Heinz, James Hone and Changgu Lee for the pristine samples preparation and Profs Kin Fai Mak and Tsachi Livneh for helpful discussions. B.R.C., C.F., L.M.M. and M.A.P. acknowledge the financial support from the Brazilian agencies CNPq, CAPES, FAPEMIG and Brazilian Nanocarbon Institute of Science and Technology (INCT-Nanocarbono). Y.W., M.T. and V.H.C. acknowledge support from the U.S. Army Research Office MURI grant W911NF-11-1-0362. Y.W. and V.H.C. acknowledge support from the NSF Materials Innovation Platform program under DMR-1539916. M.T. acknowledges support from the National Science Foundation for the grants no. 2DAREEFRI-1433311. S.M. and D.R. acknowledge financial support from the Innovation Research and Development Programme of the National Measurement System, UK, Project No. 115948 and ChemBio program on Raman Metrology. S.M. conveys thanks to Dr Barry Brennan for help in sample preparation. Open access for this article was partly funded by Physics graduate program at the Universidade Federal de Minas Gerais, Penn-State University and King’s College London.

Author contributions B.R.C., L.M.M. and M.A.P. conceived the idea and designed the experiments; B.R.C., C.F. and L.M.M. performed the resonance Raman experiments; B.R.C. analysed and interpreted the experimental data; Y.W. and V.H.C. performed the density functional theory calculation and theoretical analyses; S.M. and D.R. explored the preliminary concept and prepared the defective 1L MoS2 sample; B.R.C., Y.W. and M.A.P. wrote the paper. The work was supervised by M.T., V.H.C. and M.A.P.; All the authors discussed the results and commented on the manuscript.

Additional information Supplementary Information accompanies this paper at http://www.nature.com/ naturecommunications Competing financial interests: The authors declare no competing financial interests. Reprints and permission information is available online at http://npg.nature.com/ reprintsandpermissions/ How to cite this article: Carvalho, B. R. et al. Intervalley scattering by acoustic phonons in two-dimensional MoS2 revealed by double-resonance Raman spectroscopy. Nat. Commun. 8, 14670 doi: 10.1038/ncomms14670 (2017). Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

r The Author(s) 2017


Observing backfolded and unfolded acoustic phonons by broadband optical light scattering.

Optical polarization and intervalley scattering in single layers of MoS2 and MoSe2.

Novel Electron-Phonon Relaxation Pathway in Graphite Revealed by Time-Resolved Raman Scattering and Angle-Resolved Photoemission Spectroscopy.

Anomalous lattice vibrations of monolayer MoS2 probed by ultraviolet Raman scattering.

Immunoassay by light scattering spectroscopy.

Ice thickness measurements by Raman scattering.

Adsorption study of antibiotics on silver nanoparticle surfaces by surface-enhanced Raman scattering spectroscopy.

The structure and thermotropism of thymocyte plasma membranes as revealed by laser-raman spectroscopy.

In-situ monitoring by Raman spectroscopy of the thermal doping of graphene and MoS2 in O2-controlled atmosphere.

Dependence of coupling of quasi 2-D MoS2 with substrates on substrate types, probed by temperature dependent Raman scattering.

Drug stability analysis by Raman spectroscopy.

Confirmation of lens hydration by Raman spectroscopy.

Ciliary activity by laser light-scattering spectroscopy.

Candida parapsilosis biofilm identification by Raman spectroscopy.

Evaluation of antiatherosclerotic drugs by Raman spectroscopy.

Noninvasive glucose sensing by transcutaneous Raman spectroscopy.

Universal enantioselective discrimination by Raman spectroscopy.

Detection of Sphingomyelin Clusters by Raman Spectroscopy.

Single bacteria identification by Raman spectroscopy.

Analysis of milk by FT-Raman spectroscopy.

Characterization of coal tattoos by Raman spectroscopy.

Linear optics, Raman scattering, and spin noise spectroscopy.

Surface-enhanced Raman scattering spectroscopy of photosynthetic membranes and complexes.

Melting and premelting phenomenon in DNA by laser Raman scattering.